blohm (2005) processing of retinal and extraretinal ...wexler.free.fr/library/files/blohm (2005)...

Processing of Retinal and Extraretinal Signals for Memory-Guided SaccadesDuring Smooth Pursuit

Gunnar Blohm,1,2 Marcus Missal,2 and Philippe Lefevre1,2,3

1Centre for Systems Engineering and Applied Mechanics, Universite Catholique de Louvain, Louvain-la-Neuve, Belgium; 2Laboratory ofNeurophysiology, Universite Catholique de Louvain, Brussels, Belgium; and 3National Eye Institute, National Institutes of Health,Bethesda, Maryland

Submitted 25 May 2004; accepted in final form 10 October 2004

Blohm, Gunnar, Marcus Missal, and Philippe Lefevre. Processingof retinal and extraretinal signals for memory-guided saccades duringsmooth pursuit. J Neurophysiol 93: 1510–1522, 2005. First publishedOctober 13, 2004; doi:10.1152/jn.00543.2004. It is an essential fea-ture for the visual system to keep track of self-motion to maintainspace constancy. Therefore the saccadic system uses extraretinalinformation about previous saccades to update the internal represen-tation of memorized targets, an ability that has been identified inbehavioral and electrophysiological studies. However, a smooth eyemovement induced in the latency period of a memory-guided saccadeyielded contradictory results. Indeed some studies described spatiallyaccurate saccades, whereas others reported retinal coding of saccades.Today, it is still unclear how the saccadic system keeps track ofsmooth eye movements in the absence of vision. Here, we developedan original two-dimensional behavioral paradigm to further investi-gate how smooth eye displacements could be compensated to ensurespace constancy. Human subjects were required to pursue a movingtarget and to orient their eyes toward the memorized position of abriefly presented second target (flash) once it appeared. The analysisof the first orientation saccade revealed a bimodal latency distributionrelated to two different saccade programming strategies. Short-latency(�175 ms) saccades were coded using the only available retinalinformation, i.e., position error. In addition to position error, longer-latency (�175 ms) saccades used extraretinal information about thesmooth eye displacement during the latency period to program spa-tially more accurate saccades. Sensory parameters at the moment ofthe flash (retinal position error and eye velocity) influenced the choicebetween both strategies. We hypothesize that this tradeoff betweenspeed and accuracy of the saccadic response reveals the presence oftwo coupled neural pathways for saccadic programming. A faststriatal-collicular pathway might only use retinal information aboutthe flash location to program the first saccade. The slower pathwaycould involve the posterior parietal cortex to update the internalrepresentation of the flash once extraretinal smooth eye displacementinformation becomes available to the system.

I N T R O D U C T I O N

Space constancy is an essential feature of the visual systemthat allows us to perceive a stationary object as immobileduring self-movement even though its image shifts across theretina (Bridgeman 1995; Deubel et al. 1998; Niemann andHoffmann 1997; Stark and Bridgeman 1983). In the case wherevisual information is absent, the question arises whether spaceconstancy of memorized targets still holds during eye move-ments. This issue has been extensively studied for the saccadicsystem using the so-called double-step and colliding saccade

paradigms (Aslin and Shea 1987; Becker and Jurgens 1979;Dassonville et al. 1992; Dominey et al. 1997; Goossens andVan Opstal 1997; Hallett and Lightstone 1976a,b; Mays andSparks 1980; Mushiake et al. 1999; Schlag and Schlag-Rey1990; Schlag et al. 1989; Tian et al. 2000). In both experimen-tal conditions, a saccadic eye movement is evoked eithervisually or by microstimulation before primates have to directtheir line of sight to a previously memorized spatial location. Inthis situation, the retinal error of the memorized target does notcorrespond anymore to the required eye movement. Neverthe-less, saccades are spatially accurate. The authors conclude thatextra-retinal information about the first eye movement is avail-able to the saccadic system to adapt the second saccadeamplitude. The saccadic system is thus able to keep track of itsown movements.

During smooth-pursuit movements, the question of spaceconstancy becomes more complicated. In this case, a smootheye movement is induced before the occurrence of the saccadedirected toward the memorized target. Again, to align gazewith the correct spatial location of the memorized target, theretinal input has to be updated by extraretinal signals about thesmooth eye displacement. Thus the saccadic system needsadditional information from another motor system—thesmooth-pursuit system—to compensate for smooth eye dis-placements. Today, it is still not clear how such memory-guided saccades are programmed during smooth eye move-ments. Recent studies indicate that gaze shifts to targets mem-orized before visually guided smooth pursuit and executedafter the end of pursuit are spatially accurate (Baker et al. 2003;Herter and Guitton 1998). Also, when targets were brieflyflashed during smooth pursuit but the localization was per-formed only after the smooth-pursuit target disappeared, mem-ory-guided saccades seemed to be better predicted by thespatial error hypothesis (i.e., saccades directed to the actualtarget position in space, accounting for the retinal error andintervening movements during the latency period) than by theretinal error hypothesis (i.e., saccade directed to the retinalposition of the target irrespective of intervening eye move-ments) (Ohtsuka 1994; Schlag et al. 1990; Zivotofsky et al.1996). Taken together, these results suggest that the saccadicsystem has indeed access to information about smooth eyedisplacement during the memory period. However, when atarget is briefly flashed at the moment of the smooth-pursuittarget extinction and a targeting saccade is immediately trig-

Address for reprint requests and other correspondence: P. Lefevre, CE-SAME, Universite Catholique de Louvain, 4, Avenue G. Lemaıtre, 1348Louvain-la-Neuve, Belgium (E-mail: [email protected]).

The costs of publication of this article were defrayed in part by the paymentof page charges. The article must therefore be hereby marked “advertisement”in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

J Neurophysiol 93: 1510–1522, 2005.First published October 13, 2004; doi:10.1152/jn.00543.2004.

1510 www.jn.org

on June 8, 2008 jn.physiology.org

Dow

nloaded from

http://jn.physiology.org

gered, its amplitude is better predicted by the retinal error thanby the spatial error (Gellman and Fletcher 1992; McKenzie andLisberger 1986). These results are clearly contradictory withthe above-mentioned hypothesis of space constancy duringsmooth eye movements and the question arises what canexplain this apparent contradiction.

To answer this question, we propose to investigate howmemory-guided saccades are programmed during smooth pur-suit. In particular, because long memory periods seemed toallow space constancy during self-generated movement (Bakeret al. 2003; Herter and Guitton 1998; Ohtsuka 1994; Schlag etal. 1990; Zivotofsky et al. 1996), whereas this was not the casefor short memory periods (Gellman and Fletcher 1992; Mc-Kenzie and Lisberger 1986), we will investigate the role ofresponse latency in the saccade programming process. Bydoing this, we will be able to make the link between theabove-mentioned results of short-latency retinal and long-latency spatial saccade programming. Therefore we developeda two-dimensional (2-D) experimental paradigm in which wepresented a briefly flashed target during smooth-pursuit eyemovements. This 2-D arrangement of the paradigm allowed usto separate retinal and extraretinal signals and to obtain saccadeprogramming parameters for horizontal and vertical eye-move-ment components separately. Furthermore, a detailed analysisof saccade latencies made it possible for the first time to linkthe saccadic execution time to the programming of the mem-ory-guided eye movement. As a result, we will show that thereis a transition between retinally coded short-latency saccadesand spatially coded longer-latency eye saccades and that thistransition is reflected in a bimodal saccadic latency distribu-tion. We will also analyze which sensory parameters influencethe system’s decision to trigger short- or longer-latency sac-cades. These results shed light on the use of extra-retinalsignals when tracking smooth eye movements in the absence ofvisual input and reconcile previous contradictory results con-cerning the programming of memory-guided saccades duringsmooth self-motion. Our paradigm thus allowed us to identifytwo coupled processes for saccade execution.

M E T H O D S

Eight healthy human subjects aged between 23 and 38 yr andwithout any known oculomotor anomalies were recruited after in-formed consent. Three of them were completely naıve of oculomotorexperiments. All procedures were conducted with approval of theUniversite Catholique de Louvain Ethics Committee, in compliancewith the Helsinki Declaration (1996).

Experimental setup

Subjects faced a 1-m distant, tangent translucent screen that cov-ered about �45° horizontally and �40° vertically of their visual field.They sat in complete darkness and their head was restrained using achin rest. Two different targets—a green and a red spot— wereback-projected onto the screen. The green spot was projected by aTektronix (Beaverton, OR) 606A oscilloscope using custom opticsand measured �1.5°. This green spot was the smooth-pursuit target.A second target was a red laser spot that measured 0.2° and wasprojected onto the screen via two M3-Series mirror galvanometers(GSI Lumonics, Billerica, LA). This red spot was only briefly pre-sented during 10 ms, and we will thus refer to it as a “flash” [flashdurations between 5 and 20 ms have been used successfully (Gellmanand Fletcher 1992; Schlag et al. 1990)]. A dedicated real-time com-

puter running LabViewRT (National Instruments, Austin, TX) soft-ware controlled position, velocity, and illumination of both targets.We recorded the movements of one eye using the scleral search coiltechnique (Skalar Medical BV, Delft, The Netherlands) (Collewijn etal. 1975; Robinson 1963).

Paradigm

All recording sessions were composed of blocks of 40 trials each.Data acquisition started with two blocks of fixation control (FIX)trials followed by a varying number of blocks of test trials, such thata total recording duration of 30 min was not exceeded. Flashes werepresented during fixation (control FIX), during ongoing smooth pur-suit (“flash during ramp” FDR test) or after the pursuit ramp offset(“flash after ramp” FAR control).

Fixation control (FIX) trials started with a green central fixationspot. At a random time 500–1,000 ms after the trial began, a 10-msred flash was presented at a random position between –10° and �10°horizontally and vertically. Subjects were asked to look at the mem-orized target position as soon as the flash appeared and to fixate thisposition until the end of the trial. They had to redirect gaze to thememorized target position even though the fixation spot remainedilluminated. One thousand milliseconds after the flash, the greenfixation spot was extinguished for 500 ms to indicate the end of thetrial.

Test trials started with a 500-ms initial fixation period (green target)at 20° eccentricity (Fig. 1A, IF) in a random direction around thestraight-ahead position. The initial fixation point was thus located at arandom position on a 20° circle around the straight-ahead direction(Fig. 1A, - - -). Then the green spot performed a step away from thecenter of the screen and a ramp movement (Fig. 1A, ramp) backtoward the center of the screen. The size of the step was calculated insuch a way that the target crossed the initial fixation point after 200ms. This step was introduced to reduce the probability of occurrenceof a catch-up saccade during pursuit initiation (Rashbass 1961). Theramp velocity varied randomly between 10 and 40°/s. A 10-ms redflash (Fig. 1A, flash) was presented at a random time between 500 and1,500 ms after the ramp movement onset. The flash position wasrandomly chosen in a squared �10° (horizontal and vertical) windowaround the actual pursuit ramp position. The ramp movement contin-ued until the end of the trial. All trials lasted for 3 s. Subjects wereinstructed to follow the green pursuit target and to look at thememorized position of the flash as soon as they saw the flash.

In addition to the test trials, all subjects performed “flash afterramp” control (FAR) trials. FAR controls were similar to test trials,but the green pursuit ramp target was extinguished at a random timebetween 500 and 1,000 ms after the ramp movement onset andremained extinguished until the end of the trial. At a random time0–500 ms after the pursuit ramp extinction, a red flash was presentedin a �10° window (horizontally and vertically) around the extrapo-lated ramp position. We chose to introduce this random extinctionperiod to obtain different patterns of eye velocity after the flashpresentation. Beside this, all stimulus parameters and subject’s in-structions remained the same as for test trials.

Data acquisition and analysis

Two NI-PXI-6025E data-acquisition boards (National Instruments,Austin, TX) sampled the position of one eye and both targets (hori-zontally and vertically) at 500 Hz. Collected data were stored on ahard disk for off-line analysis with Matlab scripts (Mathworks,Natick, MA). Position signals were low-pass filtered using a zero-phase digital filter (autoregressive forward-backward filter; cutofffrequency: 50 Hz). Velocity and acceleration were derived fromposition signals using a weighted central difference algorithm on a�10-ms interval. The cutoff frequency of the derivative filter was 33 Hz.

1511MEMORY-GUIDED SACCADES DURING SMOOTH PURSUIT

J Neurophysiol • VOL 93 • MARCH 2005 • www.jn.org


Dow

nloaded from


All trials were aligned on flash onset. Figure 1B illustrates anexample in one dimension, starting from the moment of the flash onset(time: 0 ms) until 1,000 ms after the flash. The instantaneous smootheye displacement at time t was defined as the integral of the smootheye velocity EVS(t) over time, starting at the moment of the flash�SED�t� � �

0tEVs�t�dt. The smooth eye velocity EVS(t) was ob-

tained by removing saccades from the velocity trace. Saccades weredetected using a 500°/s2 acceleration threshold. We then measured theeye velocity before and after the saccades and interpolated linearlybetween these values to obtain an estimation of the smooth eyevelocity during saccades (see METHODS section of de Brouwer et al.2002 for more details).

To analyze the first saccade programming, the saccadic amplitudehad to be corrected for the contribution of the smooth-pursuit system.It has indeed been shown for horizontal eye movements that the

smooth-pursuit system does not pause during saccades and thus has asignificant contribution to the measured saccade amplitude (de Brou-wer et al. 2001, 2002). Therefore to examine the output of the saccadicsystem in isolation, the measured value of the saccade amplitude(AMP) has to be corrected by an estimate of the smooth-pursuitcontribution PAmp (see Fig. 1B). The corrected saccade amplitude isthen SAmp � AMP –PAmp. The smooth-pursuit contribution PAmp iscalculated by multiplying the saccade duration SDur with the meanvalue of the eye velocity before and after the saccade (de Brouwer etal. 2001, 2002; Keller and Johnsen 1990; Smeets and Bekkering2000). Here, we performed the same correction of the saccade am-plitude for 2-D saccades. We tested the validity of the correction foreach subject individually on the main sequence relationship betweensaccade duration SDur and vectorial amplitude AMP. For all subjects,the corrected vectorial saccade amplitude SAmp was significantlybetter correlated to SDur than the uncorrected amplitude AMP (t-test,P � 0.01). We thus validated the previously proposed method of thesaccade amplitude correction for 2-D data. All analyses in this paperwill thus use the corrected saccade amplitude SAmp.

For our analysis, we measured different parameters as illustrated inFig. 1B. The range of these parameters is provided in Table 1. At themoment of the flash, we measured the horizontal and vertical com-ponent of the position error PEF (� retinal error) and eye velocity EVF

and also the smooth-pursuit gain (gainSP,F). Following the above-described procedure, the total saccade amplitude AMP was dividedinto a purely saccadic component SAmp and a smooth-pursuit contri-bution PAmp. Position error PE(t) and smooth eye displacementSED(t) were measured continuously until 1,000 ms after the flash. Forthe analysis of the saccade latency, we partitioned our data into twodistinct subsets, i.e., foveofugal (FF) and foveopetal (FP) flashes (seeFig. 1C). In FP (FF) trials, the flash was presented ahead (behind) theactual eye position with respect to the direction of the eye trajectory.Thus as the terms “foveofugal” and “ foveopetal” intuitively suggest,the eye motion made the fovea to move either away (foveofugal) ortoward (foveopetal) the target at the moment of the flash presentation.It should be noted that a FP flash could evoke both onward andbackward saccades (with respect to the smooth eye movement direc-tion, see typical trials in Fig. 2). We also calculated the angle �R

between EVF and the position error PEF at the moment of the flash.

TABLE 1. Measured parameters

Parameter Component Mean � SD Median [25..75]%

Parameters at themoment of the flash

�PEF� (°) X 5.447 � 3.478 5.258 [2.644..7.878]Y 5.531 � 3.619 5.253 [2.584..7.997]

�EVF� (°/s) X 10.819 � 7.718 9.395 [4.687..15.231]Y 9.201 � 6.459 8.077 [4.220..12.853]

GainSP,F (.) 0.678 � 0.256 0.708 [0.500..0.865]Saccade-related

parameters�SAmp�* (°) X 5.057 � 3.524 4.506 [2.187..7.348]

Y 4.573 � 3.602 3.770 [1.670..6.716]gainS (.) X 0.908 � 0.371 0.898 [0.683..1.117]

Y 0.805 � 0.388 0.791 [0.546..1.024]�SEDS,beg� (°) X 1.992 � 1.848 1.440 [0.661..2.771]

Y 1.669 � 1.472 1.239 [0.617..2.268]�SEDS,end� (°) X 2.568 � 2.166 2.010 [0.974..3.632]

Y 2.168 � 1.724 1.742 [0.926..2.989]Final orientation

parameters�PEend� (°) X 2.319 � 2.030 1.778 [0.816..3.311]

Y 2.156 � 1.889 1.684 [0.817..2.912]�SEDend� (°) X 3.217 � 2.640 2.595 [1.219..4.503]

Y 2.851 � 2.256 2.329 [1.165..4.012]

n � 4,464 trials.

FIG. 1. Experimental paradigm and data analysis. A: test trials paradigm.The initial fixation point (E, IF), the pursuit ramp and the flash (*) arerepresented. - - -, the possible positions of the initial fixation point. B: extractedparameters. At the time of the flash (*; - - -, the memorized flash position),position error (PEF) and eye (—, saccades in bold) velocity (EVF) weresampled. The 1st saccade amplitude was divided into the purely saccadic(SAmp) and the smooth (PAmp) component. Position error (PE(t)) and smootheye displacement [SED(t)] are also represented. C: direction of smooth pursuitwith respect to the location of the flash. �R is the angle between position error(PEF) and eye velocity (EVF) at the moment of the flash. The visual field wasdivided into foveofugal (FF) and foveopetal (FP) hemifields. See text for moredetails.

1512 G. BLOHM, M. MISSAL, AND P. LEFEVRE



Dow

nloaded from


We performed our statistical analyses either using Statistica (Stat-Soft, Tulsa, OK) or Matlab (Mathworks) programs. For the presenta-tion and description of our results, we used the classical expression ofthe regression coefficient R (and corresponding P values for signifi-cance) to provide an indication of the goodness-of-fit. This was thecase for linear regressions as well as for nonlinear fitting.

R E S U L T S

We recorded a total of 5,855 test trials. All trials werevisually inspected. We discarded trials with saccades occurringaround the moment of the flash �65 ms after the flash onset(n � 956) and trials where subjects could not localize correctlythe flashed target (final error �10°) or did not trigger anysaccade (n � 435). The total number of valid trials we used inour analyses was thus n � 4,464 (�76%). We also recorded atotal of 1,957 control FIX trials and 5,919 control FAR trials,of which 1,542 (�79%) and 4,402 (�74%), respectively, werevalid. Subjects reported that they did not have any difficultiesin performing the experimental tasks.

Figure 2 shows three trials with a typical short-latency(A–C), long-latency (D–F), and very-long-latency (G–I) firstsaccade, respectively. We were only interested in saccades thatoccurred after the onset of the flash (dotted vertical line in thesix top panels of Fig. 2). The first orientation saccade in Fig. 2,A–C, had a latency of 104 ms with respect to the flash and wasalmost parallel to the position error vector at the moment of theflash PEF (� retinal error; dotted line in Fig. 2C). It seems thatthis saccade did not take into account the smooth eye move-ment during the latency period. However, a second saccadewas triggered and compensated for the remaining error. Adifferent behavior is shown in the second typical trial, in Fig.2, D–F, where the first saccade toward the flash had a latencyof 238 ms and was not parallel to PEF vector (Fig. 2F). Indeed,

this saccade seemed to take into account the fact that the eyewas moving during the latency period. However, a secondsaccade was still needed to correct for the remaining error.Figure 2, G–I, shows an extreme case where the first saccadewas triggered very late (latency: 674 ms). This example showsan almost perfect compensation for the smooth eye displace-ment that took place during the latency period. Remarkably,although the horizontal retinal position error at the moment ofthe flash was negative, the horizontal saccade amplitude waspositive. These three trials in Fig. 2 illustrate the influence oflatency on the characteristics of the first saccade. The program-ming of short latency saccades appears to be directed to theretinal position of the flash whereas there is a bias toward thespatial location of the flash when saccade latency increases.

Programming of first saccade

Typical trials in Fig. 2 seemed to indicate that short- andlong-latency saccades were not programmed in the same way.Short-latency saccades (see Fig. 2, A–C) appeared to be par-allel to the vector of retinal position error at the moment of theflash, whereas long-latency saccades (see Fig. 2, D–I) indicatethat extra-retinal signals about the ongoing eye movementduring the latency period of the saccade could also influence itsprogramming. Long-latency saccades would thus be spatiallymore accurate than the retinal short-latency saccades. Figure 3shows a comparison of the retinal versus the spatial saccadicprogramming hypothesis as a function of saccade latency.Figure 3 shows scatter plots of the first saccade direction forthe retinal and spatial error hypothesis, for short (�150 ms)-and long (�300 ms)-latency trials separately. This confirmedthe tendency shown in our example trials (Fig. 2), i.e., short-latency saccades were better described by the retinal error

A D G

C F I

B E H

FIG. 2. Typical trials. Three typical trialsare shown. A–C: a short-latency (104 ms)trial; D–F: a long-latency (238 ms) trial; andG–H: very-long-latency (674 ms) trial. A, D,G: position vs. time representation of thepursuit target (dashed lines), the flash (hor-izontal dotted lines stand for the memorizedflash position) and the movement of 1 eye(solid lines) separately for horizontal (red,H) and vertical (blue, V) components. Rel-evant saccades are represented with boldlines. The vertical dotted line indicates whenthe flash was presented. B, E, H: velocity vs.time representation of the trial. Saccades arerepresented with thin lines; other conven-tions are the same as in A, D, and G. C, F, I:vertical vs. horizontal representation of thetrial. The dashed line represents the pursuitramp target and the arrow indicates themovement direction. The eye position is rep-resented with a red dot every 6 ms. When thedots are separated, this indicates a fast (sac-cadic) eye movement; otherwise the move-ment is smooth. The flash (star) is connectedto the eye trace by a thin dotted line thatindicates where the eye was at the momentof the flash (� PEF).




Dow

nloaded from


hypothesis, whereas long-latency saccades were spatially moreaccurate. A qualitatively and quantitatively very similar behav-ior was observed for FAR control trials (data not shown).

To analyze the transition between the retinal and spatialprogramming of the first orientation saccade, we developed asmooth eye displacement compensation index (CI). This indexvaries from zero (retinal error hypothesis) to one (spatialsaccade programming) and is calculated as follows

CI � 1 �PE�

SED(1)

Position error (PE�) and smooth eye displacement (SED) weremeasured in the direction of (and thus parallel to) smoothpursuit. Following Eq. 1, if PE� � �SED, then the system didnot compensate for SED and thus CI � 0 (retinal program-ming). If the system does compensate for SED, then PE� � 0and thus CI � 1 (spatial programming). Values for CI between0 and 1 indicate the percentage of SED compensation of asaccade. Figure 4 shows the results of this analysis. Figure 4Ashows the distribution of the compensation index CI for short(�150 ms, 1) and long (�300 ms, �) latency first saccades. InB, we represented the evolution of CI with saccade latency.This shows that short-latency saccades were �85% retinal andonly 15% spatial. In contrast, the programming for longer-latency saccades converges �75% spatial coding, leaving a25% retinal contribution to the saccadic programming. Further-more, Fig. 4C shows the evolution of the compensation indexover time for each subject individually. This confirms theresults from Fig. 3 indicating that short-latency saccades areprogrammed retinally, whereas there is a transition toward thespatial error hypothesis for longer saccade latencies.

To be sure that the retinal to spatial transition we observedwas not due to some side-effect of saccade programming, wemeasured the saccadic gain orthogonal to the pursuit ramp

movement direction in Fig. 5. Because there was no SED inthis orthogonal direction, the saccadic gain was a good mea-sure of the movement’s accuracy. As a result, there was neithera significant modulation of this saccadic gain for differentlatencies, nor a difference between test trials and FIX controltrials (see Fig. 5). It should be mentioned that the scatter in Fig.5 was essentially due to small position errors.

Saccadic latency distribution

We plotted the first saccade latency histogram for FDRtrials, FAR controls, and FIX controls in Fig. 6, A–C, respec-tively. One can easily identify two modes in the latencydistribution. To further characterize the two modes of thesehistograms, we fitted a double-recinormal distribution to ourdata. The double-recinormal distribution for the latencies hasthe following expression

1

Tlat

� A1 � gauss��1,�1� � A2 � gauss��2,�2� (2)

The inverse of the saccade latency Tlat was thus fitted by twoindependent Gaussians. This is equivalent to the hypothesis oftwo separate and noninteracting decision processes of theLATER type (Carpenter and Williams 1995). Indeed, theLATER model states that the inverse of saccade latency isproportional to the rate of rise of a decision signal divided bythe decision threshold. The particularity of the LATER modelis that the decision signal is assumed to rise linearly with anormally distributed rate of rise. We essentially chose thisdouble-recinormal fit function because it described well ourdata and allowed us to quantify with very few parameters theentire latency distribution. However, to justify the use of thisparticular probability density function, we performed a k-foldcross-validation applying the Kolmogorov-Smirnov statistical

FIG. 3. Retinal vs. spatial saccade programming. Scatter plotsof raw data for the 1st saccade direction (SDir) as a function of thedirection of the position error (PEDir). The retinal error hypoth-esis (A and B) is compared with the spatial error hypothesis (Cand D) separately for short-latency trials (latency �150 ms, Aand C) and long-latency trials (latency �300 ms, B and D).




Dow

nloaded from


one-sample distribution test. Therefore we used a randomsubset of 75% of our data to estimate the fit parameters of Eq.2 by means of standard least-square data fitting using theGauss-Newton method. Afterward, we performed a Kolmog-orov-Smirnov analysis to test if the remaining 25% of our datawere distributed identically to the previously identified distri-bution. This procedure was performed k � 1,000 times. As aresult, we found an average 98.3% acceptance of the double-recinormal probability density function hypothesis (5% signif-icance level).

To estimate the fit parameters, we applied the same least-square data fitting as used above. As a result, we obtained themeans and SDs (inverse time scale) for the best fit of Eq. 2 onthe histograms. Hereafter, we will provide values for mi � 1/�i

(location of the maximum) and si �1

2� � 1

�i � �i

�1

�i � �i�

(estimated scatter), which are in the real-time domain and thusintuitively easier to interpret. Maxima � scatter were 115 � 40and 225 � 41 ms, respectively, for both modes of the double-recinormal fit (dotted line) on the latency histogram of FDRtrials in Fig. 6A, 110 � 35 and 222 � 25 ms for FAR trials inFig. 6B, and 134 � 36 and 220 � 21 ms for FIX controls inFig. 6C. Note, that there were also two distinct latency modesfor both FAR and FIX control trials. However, in the FARcontrol situation, the less frequent second mode might be dueto the fact that the flashed stimulus did not compete any morewith the pursuit ramp target. Interestingly, the minimum be-tween the two modes in Fig. 6A was located �175 ms, whichwas approximately the time needed for the extraretinal smootheye movement signal to be taken into account for the saccadeprogramming (see CI in Fig. 4). The individual latency histogramsof FDR trials for each subject in Fig. 6D demonstrate that bothlatency modes were present in each subject, although with differ-ent proportions and slightly different locations of the maxima.

Previous studies showed that there is an inhibition of sac-cade initiation to previously attended positions (Klein 2000)

and that there is a directional asymmetry of saccade latencyduring smooth-pursuit eye movements (Kanai et al. 2003;Tanaka et al. 1998), i.e., saccades executed in the same direc-tion as pursuit have shorter latencies than saccades in theopposite direction. All these experiments describe saccadelatencies to visible stationary or moving targets. Here weinvestigated whether a similar behavior could also be observedfor memory-guided saccades to briefly flashed targets. Further-more, previous studies that analyzed smooth-pursuit-relateddirectional differences in saccade latency used only horizontalstimuli and eye movements. Here, our 2-D paradigm will allowus to test saccadic latencies for different positions of the flashwith respect to the smooth-pursuit direction.

To address the question of a possible directional asym-metry of saccade latencies, we separated our data intofoveofugal (FF) and foveopetal (FP) trials (see METHODS

section, Fig. 1C). Figure 7, A and C, shows the latencydistributions of FF (1, - - -) and FP (�, � � � ) data separately for

FIG. 5. Saccade accuracy. The normalized saccade amplitude perpendicu-lar to the initial ramp movement direction was used as an indicator for thesubject’s saccade accuracy. The latency-dependent computations of this me-dian saccade gains were compared with the fixation (FIX) control situation. U,individual saccades; ■ and �, the median and quartiles. In addition, individualtests for each subject showed no significant latency dependence of the saccadeaccuracy (t-test, P � 0.1).

FIG. 4. First saccade compensation indexfor smooth eye movements. A: histograms forthe distribution of the compensation index ascalculated in Eq. 1 for short-latency trials(latency �150 ms, 1) and long-latency trials(latency �300 ms, �). A 0 compensationindex corresponds to retinal saccade pro-gramming, whereas a unitary compensationindex corresponds to spatially accurate sac-cades. ‚ and Œ refer to B. Histograms arenormalized in amplitude. B: evolution of thecompensation index for different saccade la-tencies. ■ and whiskers indicate means � SE(25-ms bins). *, when the means are notsignificantly different from 0 (t-test, P �0.05). 1 with Œ: corresponds to the datarepresented in 1 of A. � with ‚, the datarange of � in A. C: evolution of the compen-sation index for different saccade latenciesindividually for each of the 8 subjects (50-msbins).




Dow

nloaded from


FDR and FAR trials. Interestingly, the relative importance ofthe short- and long-latency modes changed between FP and FFtrials, whereas their location was approximately constant. FPtrials contained more short-latency than long-latency saccades,whereas the opposite was the case for FF trials. Thus the meanlatencies for saccades to the FF and FP hemifield were 214 and191 ms, respectively for FDR trials (Fig. 7A) and 158 and 172ms, respectively for FAR controls (Fig. 7C). The differencebetween both means was highly significant (t-test, P � 0.0001)in both cases. Note that the FP and FF data sets were slightlydifferent in their total number of trials; this was due to anasymmetry in the paradigm. Therefore we also tested theFP/FF hemifield difference for a subset of our data withidentical properties (we used the same range and distribution ofPEF) for FP and FF and found no difference with the above-reported results. On average, the latency of FF saccades waslarger, but they were spatially more accurate when comparedwith FP saccades.

Our 2-D paradigm allowed us to go one step further and toask for the first time whether the asymmetry in saccadelatencies reported in Fig. 7, A and C, was due to a pursuitrelated focus of attention or whether it might be the result of aninhibitory hemifield effect. Therefore we present in Fig. 7, Band D, polar plots of mean saccade latencies depending on theangle �R between the flash position and the pursuit eye-movement direction at the moment of the flash (see METHODS

section, Fig. 1C). It can be easily observed that almost all meanlatencies within the same hemifield had the same values andthat there was a relatively sharp transition between the FP andFF hemifields.

How did the system decide whether to rapidly trigger a shortlatency but inaccurate saccade or to wait longer and trigger aspatially more accurate saccade? We investigated a possibledependence of the first saccade latency on the main sensoryparameters measured at the moment of the flash appearance,i.e., the position error PEF and eye velocity EVF at the momentof the flash. In addition, we tested other parameters, like thesmooth-pursuit gain (gainSP,F) at the moment of the flash, thetarget velocity or the duration of ongoing smooth-pursuit eyemovement, but the overall regression results were best usingthe above-mentioned sensory variables. This analysis was alsomotivated by previous findings that showed a dependence ofthe mean saccade latency on the distance between the eye andthe target (Bell et al. 2000; Clark 1999; Hodgson 2002;Kalesnykas and Hallett 1994).

The position error PEF at the moment of the flash was thefirst sensory parameter that influenced the distribution of sac-cade latency. Figure 8A shows the dependence of the firstsaccade latency on the distance PEF between the flash and theeye for FDR (black) and FAR (gray) data. Mean values andstandard errors in Fig. 8A essentially indicated that for smallposition errors (PEF� 5°) the latency was larger than for largerPEF. We separated our data into three categories depending onPEF, i.e., small (PEF� 5°), medium (5°� PEF� 10°) and large(10°� PEF) values, and computed the latency histograms anddouble-recinormal fits in Fig. 8B. For small PEF values, thelong latency mode of the histogram was more important thanfor larger PEF values. While the value of the location of theshort-latency mode’s maximum remained roughly constant, the

FIG. 6. First saccade latency. A: latencyhistogram for flash during ramp (FDR) trials(n � 4,464). The dotted line represents adouble-recinormal fit function with maxima at115 and 225 ms. B: latency histogram forflash after ramp (FAR) control trials (n �4,402). The double-recinormal fit function hasmaxima at 110 and 222 ms. C: latency distri-bution for FIX control trials. Fitted distribu-tion maxima are at 134 and 220 ms. D: indi-vidual histograms for the 1st saccade latencydistribution of all 8 subjects in the case ofFDR trials. The fitted distribution maximavaried between 96 and 163 ms for the 1stlatency mode and between 207 and 286 msfor the 2nd mode.




Dow

nloaded from


longer latency mode of the distribution moved toward shorterlatencies with increasing PEF.

The second sensory parameter we found to influence the firstsaccade latency was the smooth eye velocity EVF at themoment of the flash. Figure 8C shows that the saccade latencydepended approximately linearly on EVF for FDR (black) andFAR (gray) data. The regression equation on raw data of thelinear fit (solid line) was y � 158 ms � 2.48 ms* x (R � 0.183,P � 0.0001) for FDR trials and y � 139 ms � 2.80 ms* x (R �0.226, P � 0.0001) for FAR controls. Similar to Fig. 8B, wesubdivided our dataset into different ranges of the sensoryparameter EVF (vertical dotted lines in Fig. 8C) and plotted thecorresponding latency histograms and double-recinormal fits inFig. 8D. Figure 8D clearly demonstrates that the shift in meansaccade latency with higher EVF values was due to the in-creased relative importance of the long-latency mode. Thiseffect was underlined by the fact that the location of thelong-latency mode’s maximum was slightly shifted towardlarger values with increasing EVF.

As a conclusion, our results concerning the influence of thesensory parameters on the first saccade latency revealed twodistinct effects. First, large eye velocity (EVF) and smallposition error (PEF) increased the relative importance of thelong-latency mode with respect to the shorter-latency mode.Second, we observed a shift of the long-latency mode towardlarger values when eye velocity (EVF) was high and positionerror (PEF) was small. Performing the same analysis using bothsensory parameters in combination increased this effect yield-ing to an absence of the first latency mode for combined smallPEF and large EVF, whereas long-latency saccades disappearedfor a combination of large PEF and small EVF (data notshown).

We would like to emphasize that the hemifield differencebetween FP and FF flashes (see Fig. 7) was still present in allanalyses concerning the sensory parameters in Fig. 8 (data notshown) and significant (t-test, P � 0.01). We explicitly testedthat our results were not an artifact of some combined param-eter effect or even due to slight asymmetries of our dataset. Asa result, we report here that both the FP/FF latency differenceand the latency dependence on the sensory parameters at thetime of the flash were consistent and unbiased effects. Further-more, even though there were differences in the individuallatency histograms when computed for each subject separately, allthe above-described latency effects were present for all subjects.

Time course of orientation

After the first orientation saccade toward the memorizedposition of the flash, we usually observed one or more correc-tive saccades that brought the eye closer to the spatial positionof the flash. This was the case irrespectively of whether the firstsaccade was triggered with short or longer latency (see typicaltrials in Fig. 2). The way in which the system reaches thememorized goal determined by a flash during smooth eyemovements has previously been investigated for horizontal eyemovements by Blohm et al. (2003), who showed that thesaccadic system was able to compensate for smooth anticipa-tory eye movements with a certain delay. Here, we observedqualitatively the same behavior, i.e., some time was needed forthe smooth eye displacement to be taken into account (see 1stsaccade programming results).

As already mentioned, the purpose of secondary “catch-up”saccades was to compensate for the remaining uncorrectedsmooth eye displacement. Thus to analyze in more details the

FIG. 7. Saccade latency dependence on direc-tion. A: The latency histogram for FDR trials as inFig. 6A but divided into 2 populations, i.e., fove-ofugal (FF, 1, n � 1,884) and foveopetal (FP, �,n � 2,580) flash presentations. Maxima for thelatency fit on the data were 129 and 223 ms for FF(- - -) and 110 and 228 ms for FP ( � � � ) flashes. B:polar plot of FDR saccade latencies as a functionof the angle of flash appearance �R relative to thesmooth pursuit direction. �R � 0° corresponds toa flash presented straight ahead in the direction ofthe eye movement. E and whiskers indicatemean � SE for the 10°-angle bins. � � � , meanvalues of the FF (214 ms) and FP (191 ms)hemifield separately. The difference between bothmeans was highly significant (t-test, P � 0.0001).C: as for A, the FAR latency histogram wasdivided into foveofugal (FF, 1, n � 1,768) andfoveopetal (FP, �, n � 2,634) flash presentations.FF maxima were at 113 and 207 ms; FP maximawere at 109 and 228 ms. D: polar plot of saccadelatencies as a function of the flash direction rela-tive to the smooth pursuit direction. The sameconventions as in B apply. The mean values of FFand FP hemifield saccade latencies were 172 and158 ms, respectively. The difference between bothmeans was highly significant (t-test, P � 0.0001).




Dow

nloaded from


consequences of the secondary saccades, we computed thecompensation index (CI) of Eq. 1 for up to four saccades afterthe flash presentation and for different saccade latencies (mea-sured with respect to the flash onset). The results of thisanalysis are shown in Fig. 9 for both FDR trials (black) andFAR controls (gray). Note that there is an initial increase of thecompensation index only for the first saccade (shown in moredetails for FDR trials on Fig. 4). For all successive saccades,this index is approximately constant. It should also been notedthat the three secondary catch-up saccades did not completelycompensate for the smooth eye displacement.

Finally, we analyzed the contribution of the total smooth eyedisplacement SEDend to the final position error PEend, takinginto account that there might also be a residual contribution ofthe retinal position error PEF (measured at the moment of theflash) to PEend. Table 1 indicates the ranges of these parame-ters. The second-order regression for PEend with the variablesPEF and SEDend provided the following results for the hori-zontal and vertical components of FDR trials

PEend,X � (�.121 � .037) � �.013 � .006��PEF,X

� �.465 � .009��SEDend,X �R � 0.622, P 0.001� (3)

PEend,Y � (�.703 � .035) � �.091 � .005��PEF,Y

� �.429 � .010��SEDend,Y �R � 0.559, P 0.001� (4)

As a consequence, the saccadic system compensated differ-ently for the retinal error information (PEF) and the extraretinal

information about the total smooth eye displacement (SEDend).The compensation for the retinal flash position error PEF was98.7% (subject variability: 80.1–121.1%) horizontally (Eq. 3)and 90.9% (subject variability: 71.3–106.5%) vertically (Eq. 4)at the end of the orientation process. Note that this was a betterperformance than after the first orientation saccade (see Fig. 5and Table 1). Furthermore, the final compensation for thesmooth eye displacement (extraretinal signal) was 53.4% (sub-ject variability: 36.5–87.2%) horizontally (Eq. 3) and 57.1%(subject variability: 45.0–73.3%) vertically (Eq. 4). Thus Eqs.3 and 4 provided a measure of the accuracy of the orientationprocess that compensated almost perfectly for PEF and ac-counted for �55% of SED (see compensation index, Fig. 9,B–D). It is worth noting that we computed the compensationindex only if a saccade took place, whereas the regressionanalysis of Eqs. 3 and 4 included all trials irrespectively of thenumber of executed saccades. Our values were lower than thepreviously reported 70% overall SED compensation in the caseof target localization during smooth anticipatory eye move-ments (Blohm et al. 2003). The regression analysis for FARcontrol provided results similar to the test trials (data not shown).

D I S C U S S I O N

The programming of memory-guided saccades duringsmooth eye movements reflects the performance of the systemin maintaining space constancy. To investigate this mecha-

FIG. 8. Sensory parameters influence saccadelatency. A: 1st saccade latency as a function ofthe distance PEF between the flash and the eyes.black box, FDR trials; white box, FAR controls.Squares and whiskers indicate mean and SE(n � 27–493). Separations (vertical dotted lines)of data for values of PEF refer to histograms inB. B: latency histograms (solid lines) and dou-ble-recinormal fits (dotted lines) for 3 subsets ofdata with respect to the range of PEF values andfor FDR (black) and FAR (gray) trials sepa-rately. C: 1st saccade latency dependence on theeye velocity EVF at the moment of the flash.Black data are FDR trials; gray data stand forFAR controls. Squares and whiskers indicatemean and SE (n � 49 to 1225). The solid linerepresents the linear fit on raw data. Verticalseparations (dotted lines) refer to panel D. D:latency histograms (solid lines) and double-reci-normal fits (dotted lines) for different subsets ofthe data with respect to different ranges of EVF

values and for FDR (black) and FAR (gray)trials separately.




Dow

nloaded from


nism, we presented a briefly flashed target during smooth-pursuit eye movements and analyzed the characteristics of thefirst saccade. As a result, we found that short-latency saccadeswere better correlated with retinal error than with spatial error,whereas the opposite was the case for longer-latency saccades.The saccadic system approximately needed 175 ms beforeextraretinal information about the smooth eye displacementcould be used in saccade programming. This behavior was alsoreflected in the latency distribution, where we found twodistinct modes—a short-latency and a longer-latency mode—separated at �175 ms. We interpret our results as evidence forthe existence of two different neural processes for saccadeprogramming: one fast but inaccurate and the other slower butspatially more accurate.

Saccadic reaction times

We observed a bimodal distribution of saccade latencies inour test trials and in both control data sets. Similar distributionshave been observed previously during saccadic “gap” and“overlap” paradigms (Fischer et al. 1993, 1997; Reulen1984a,b; Saslow 1967). To quantify our bimodal first saccadelatency distribution, we fitted a double-recinormal distributionto our data. We used this particular distribution because itdescribed very well our data with only very few parameters.However, as already mentioned in RESULTS, this procedure istheoretically equivalent to the hypothesis of two distinct sac-cade trigger mechanisms of the LATER type (Carpenter andWilliams 1995; Reddi and Carpenter 2000; Reddi et al. 2003)that work in parallel and do not interact. Although we cannotprove that such a mechanism exists, it is supported by our databecause all latency distributions agree with the hypothesis oftwo parallel decision processes. As we demonstrated in Fig. 8,the system’s decision between both processes depended on theset of sensory parameters (distance of the target from the fovea

and eye velocity) at the moment of the flash. Indeed, the effectof position error on saccade latency has previously beendescribed (Bell et al. 2000; Clark 1999; Hodgson 2002; Kale-snykas and Hallett 1994), and it makes sense that the systemneeds more time for small position errors to decide whether itis necessary to trigger a saccade. We did also observe thiseffect for our fixation control trials (data not shown) and our“flash after ramp” (FAR) controls (Fig. 8, A and B), i.e., thelonger latency mode essentially resulted from flashes presentedat small position errors. The influence of eye velocity onsaccade latency however was—to our knowledge—a novelfinding. Because for large smooth-pursuit velocities the errorafter a short latency saccade (retinally programmed) would bebig, the system might prefer to wait longer for extraretinalsmooth eye displacement information to become available (seeFig. 4). This “waiting strategy” allowed the system to performspatially more accurate initial saccades. Thus the relationshipbetween saccade latency and eye velocity described the sys-tem’s tradeoff between speed and accuracy.

We showed in Fig. 7 that the mean saccadic latency isshorter for flashes presented in the direction of the movement(foveopetal) than for flashed targets presented in the oppositedirection (foveofugal). This is compatible with previous find-ings for horizontal catch-up saccades during smooth pursuit ofcontinuously visible targets (Kanai et al. 2003; Tanaka et al.1998). In these studies, saccades in the same direction assmooth pursuit had shorter latencies than saccades in theopposite direction. In addition, this latency asymmetry wasalso present in our FAR control trials where no visible targetwas present. Kanai et al. (2003) hypothesized that this differ-ence in latencies is due to the inhibition of saccades topreviously attended positions during smooth pursuit, a partic-ular instantiation of the “inhibition of return” (IOR) effect (seeKlein 2000 for a review). At first sight, our data seemed to becompatible with such a hypothesis. However, our 2-D para-

FIG. 9. Consequences of the secondary“catch-up” saccades. Compensation indices asa function of saccade latency with respect tothe onset of the flash. Black data are FDRtrials; gray data stand for FAR controls.Squares and whiskers indicate mean and SE.A: compensation index for the 1st saccade (forFDR: the same data as in Fig. 4B but withlarger bin sizes). B: compensation index of the2nd saccade (n � 3,378). C: effect of the 3rdsaccade on the compensation index (n �1,743). D: compensation index for the 4thsaccade (n � 707).




Dow

nloaded from


digm allowed us for the first time to show that this latencyasymmetry is a hemifield effect and not related to a “focus ofattention,” as IOR would predict (Maylor and Hockey 1985;Posner et al. 1985). Nevertheless, our results would be com-patible with an attentional facilitation in the direction of themovement, if this facilitation was extended to the wholefoveopetal hemifield and was not restricted to a focus ofattention (Maylor and Hockey 1985; Posner et al. 1985). Thepurpose of such a hemifield bias could simply be to facilitatemovement in the direction of heading.

Saccades compensate for self-motion

The analysis of the saccade latencies revealed a bimodaldistribution. The presence of these two different saccade trig-ger processes was also reflected in the way saccades wereprogrammed. We showed indeed that saccades with shortlatencies (�175 ms) were programmed using the only avail-able retinal information, i.e., position error at the moment ofthe flash. This is in accordance with previous studies (Gellmanand Fletcher 1992; McKenzie and Lisberger 1986) and con-trasts with the situation where continuous visual feedback ispresent. In the latter condition, orientation saccades to theobject’s spatial location are programmed using two types ofretinal information, i.e., position error and the relative velocityof the eyes with respect to the target (retinal slip) (de Brouweret al. 2001, 2002; Gellman and Carl 1991; Keller and Johnsen1990; Ron et al. 1989). However, although retinal informationabout smooth self-motion was absent in our experiment, longer-latency (�175 ms) saccades used a different programmingmechanism that included extraretinal information about thesmooth eye displacement in addition to the retinal positionerror.

All previous studies support our results. McKenzie andLisberger (1986) reported that short-latency memory-guidedsaccades during pursuit eye movements were retinally coded.However, one monkey showed a bias of the suggested retinallysaccade coding toward a more accurate spatial amplitudeprogramming, that the authors attributed to the monkey’sparticipation in previous smooth tracking experiments. How-ever, this monkey also had particularly long mean saccadereaction times compared with the two other monkeys. Webelieve that our results explain the third monkey’s bias towardthe spatial coding hypothesis simply as a consequence oflonger latencies. Other studies showed that when memory-guided saccades were executed after a delay period, smoothself-motion was compensated and saccades were spatiallyaccurate (Baker et al. 2003; Herter and Guitton 1998; Ohtsuka1994; Schlag et al. 1990; Zivotofsky et al. 1996), which is fullycompatible with our results. Thus we described here the miss-ing piece in the puzzle of self-motion integration, i.e., thetemporal transition between the retinal and spatial representa-tion of the visual world in the oculomotor system.

Initially, McKenzie and Lisberger (1986) used a paradigmsimilar to our FAR controls in an attempt to differentiatebetween two different types of saccade models, i.e., positionand displacement models. Position models (Robinson 1975;Van Gisbergen et al. 1981) assume that the signal of desiredeye position is compared with a signal of the current positionof the eye in the orbit to generate a motor command. Incontrast, displacement models (Jurgens et al. 1981) assume that

a desired displacement—rather than a desired position—signalis used to drive the saccadic eye movement. During a saccade,this desired displacement is compared with an internal repre-sentation of the movement already executed to produce themovement command. McKenzie and Lisberger (1986) trainedmonkeys to make saccades to flashes memorized duringsmooth eye movements to test the position versus the displace-ment model. The monkeys made saccades to the retinal posi-tion of the flash (Gellman and Fletcher 1992; McKenzie andLisberger 1986), which validated the displacement model be-cause the position model would have predicted a spatiallyaccurate eye movement. However, our data reported here aswell as previous findings (Baker et al. 2003; Herter and Guitton1998; Ohtsuka 1994; Schlag et al. 1990; Zivotofsky et al. 1996)support the idea that if more time is available to the systembefore the onset of the orienting eye movement, saccades canbe spatially accurate. This implies that there must be anadditional mechanism available to perform this retinal to spa-tial transformation, but apparently, such a mechanism takessome time and is not implemented at the level of the saccadicdisplacement integrator. Hereafter, we will propose a neuralmechanism that could account for all the data.

Hypothesized underlying neural mechanisms

We believe that our results reflect the presence of twodifferent neural mechanisms for retinal and extraretinal infor-mation processes characterized by different processing times.In this section, we will shortly lay out one hypothesis of theunderlying neural pathways that might be involved in theintegration of smooth-pursuit eye movements and the orienta-tion to memorized targets.

The major forebrain and midbrain structures involved in theprogramming and/or control of saccades are the posteriorparietal cortex (PPC), the frontal eye fields (FEFs), the dorso-lateral prefrontal cortex (DLPC), the basal ganglia, the cere-bellum (CB), and the superior colliculus (SC) (see for a reviewKrauzlis and Stone 1999; Leigh and Zee 1999). It is generallyaccepted that SC is essential to generate short-latency saccades(Fischer and Ramsperger 1986; Munoz and Wurtz 1992;Schiller et al. 1987). Indeed, Schiller at al. (1987) showed alateralized absence of short-latency saccades in monkeys afterunilateral SC ablation, whereas longer-latency saccades werestill present. This was not the case for FEF ablation, which hadno long-term effect on saccade latencies (Schiller et al. 1987).Therefore we propose that the short latency saccades wereported here were essentially mediated by a fast “striatal-collicular pathway” (Leigh and Zee 1999). This contrasts withlonger-latency saccades that are known to involve most of theabove-cited structures including PPC.

It has previously been suggested (Duhamel et al. 1992;Heide et al. 1995), that PPC plays a key role to account forextraretinal signals (Tobler et al. 2001) when keeping track ofself-motion to ensure space constancy. Indeed, the lateralintraparietal area (LIP) and area 7a in PPC receive informationabout upcoming saccades to update the spatial representationof visual stimuli (Andersen et al. 1985; Bremmer et al. 1997).Therefore we propose that the memorized flash position in ourexperiment is stored in PPC (Barash et al. 1991; Pare andWurtz 1997; Pierrot-Deseilligny et al. 1991) and updated bysmooth eye displacement information when it becomes avail-




Dow

nloaded from


able. The smooth eye displacement signal could result from anintegration process of the smooth motor command (Blohm etal. 2003), which might take some time. Such a process couldinvolve parts of the cerebellum—highly involved in generatingsmooth pursuit (Lisberger et al. 1987; Pola and Wyatt 1991)—and smooth eye displacement information could be projectedeither directly or via the thalamus (Clower et al. 2001) to PPCto update the spatial representation of the flash. This wouldexplain why in our data smooth eye displacement informationtook �175 ms to become available to the saccadic system.According to this view, the classical saccade pathway would beresponsible for retinally coded saccades. An additional feed-back pathway for the integration of smooth-pursuit eye move-ments could then be added to these classical structures toensure space constancy during smooth eye movements. Ourobservation that smooth eye movement integration was char-acterized by a delay (�175 ms) implies that the system onlycompensates for the smooth eye displacement that has alreadybeen integrated. This might explain why the compensationmechanism is not an all-or-none process. However, we cannotexclude a possible role of proprioception (Steinbach 2000) forthe spatial orientation toward the flash, although propriocep-tion is thought not to be a predominant source for spatiallocalization (Weir 2000) and the control of eye movements ingeneral (Lewis et al. 2001).

We suggest using our original 2-D paradigm in neuralrecording studies to identify the neural correlates underlyingthe monitoring of smooth pursuit. The presence of two separatecontrol modes for memory-guided saccades during smooth eyemovements could provide a new testing bench to investigatethe neural processes of smooth motion integration. More gen-erally, our results provide a new path to investigate the inter-action between smooth pursuit and saccadic systems.

A C K N O W L E D G M E N T S

We thank the anonymous referees for very helpful suggestions concerningthe presentation of our data. The scientific responsibility rests with its authors.

G R A N T S

This work was supported by the Fonds National de la Recherche Scienti-fique; the Fondation pour la Recherche Scientifique Medicale; the Belgianprogram on Interuniversity Attraction Poles initiated by the Belgian FederalScience Policy Office; and internal research grant (Fonds Speciaux de Recher-che) of the Universite catholique de Louvain.

R E F E R E N C E S

Andersen RA, Essick GK, and Siegel RM. Encoding of spatial location byposterior parietal neurons. Science 230: 456–458, 1985.

Aslin RN and Shea SL. The amplitude and angle of saccades to double-steptarget displacements. Vision Res 27: 1925–1942, 1987.

Baker JT, Harper TM, and Snyder LH. Spatial memory following shifts ofgaze. I. Saccades to memorized world-fixed and gaze-fixed targets. J Neu-rophysiol 89: 2564–2576, 2003.

Barash S, Bracewell RM, Fogassi L, Gnadt JW, and Andersen RA.Saccade-related activity in the lateral intraparietal area. II. Spatial properties.J Neurophysiol 66: 1109–1124, 1991.

Becker W and Jurgens R. An analysis of the saccadic system by means ofdouble step stimuli. Vision Res 19: 967–983, 1979.

Bell AH, Everling S, and Munoz DP. Influence of stimulus eccentricity anddirection on characteristics of pro- and antisaccades in non-human primates.J Neurophysiol 84: 2595–2604, 2000.

Blohm G, Missal M, and Lefevre P. Interaction between smooth anticipationand saccades during ocular orientation in darkness. J Neurophysiol 89:1423–1433, 2003.

Bremmer F, Distler C, and Hoffmann KP. Eye position effects in monkeycortex. II. Pursuit- and fixation-related activity in posterior parietal areas LIPand 7A. J Neurophysiol 77: 962–977, 1997.

Bridgeman B. A review of the role of efference copy in sensory andoculomotor control systems. Ann Biomed Eng 23: 409–422, 1995.

Carpenter RH and Williams ML. Neural computation of log likelihood incontrol of saccadic eye movements. Nature 377: 59–62, 1995.

Clark JJ. Spatial attention and latencies of saccadic eye movements. VisionRes 39: 585–602, 1999.

Clower DM, West RA, Lynch JC, and Strick PL. The inferior parietallobule is the target of output from the superior colliculus, hippocampus, andcerebellum. J Neurosci 21: 6283–6291, 2001.

Collewijn H, van der Mark F, and Jansen TC. Precise recording of humaneye movements. Vision Res 15: 447–450, 1975.

Dassonville P, Schlag J, and Schlag-Rey M. The frontal eye field providesthe goal of saccadic eye movement. Exp Brain Res 89: 300–310, 1992.

de Brouwer S, Missal M, Barnes G, and Lefevre P. Quantitative analysis ofcatch-up saccades during sustained pursuit. J Neurophysiol 87: 1772–1780,2002.

de Brouwer S, Missal M, and Lefevre P. Role of retinal slip in the predictionof target motion during smooth and saccadic pursuit. J Neurophysiol 86:550–558, 2001.

Deubel H, Bridgeman B, and Schneider WX. Immediate post-saccadicinformation mediates space constancy. Vision Res 38: 3147–3159, 1998.

Dominey PF, Schlag J, Schlag-Rey M, and Arbib MA. Colliding saccadesevoked by frontal eye field stimulation: artifact or evidence for an oculo-motor compensatory mechanism underlying double-step saccades? BiolCybern 76: 41–52, 1997.

Duhamel JR, Colby CL, and Goldberg ME. The updating of the represen-tation of visual space in parietal cortex by intended eye movements. Science255: 90–92, 1992.

Fischer B, Gezeck S, and Hartnegg K. The analysis of saccadic eyemovements from gap and overlap paradigms. Brain Res Brain Res Protoc 2:47–52, 1997.

Fischer B and Ramsperger E. Human express saccades: effects of random-ization and daily practice. Exp Brain Res 64: 569–578, 1986.

Fischer B, Weber H, Biscaldi M, Aiple F, Otto P, and Stuhr V. Separatepopulations of visually guided saccades in humans: reaction times andamplitudes. Exp Brain Res 92: 528–541, 1993.

Gellman RS and Carl JR. Motion processing for saccadic eye movements inhumans. Exp Brain Res 84: 660–667, 1991.

Gellman RS and Fletcher WA. Eye position signals in human saccadicprocessing. Exp Brain Res 89: 425–434, 1992.

Goossens HH and Van Opstal AJ. Local feedback signals are not distortedby prior eye movements: evidence from visually evoked double saccades.J Neurophysiol 78: 533–538, 1997.

Hallett PE and Lightstone AD. Saccadic eye movements to flashed targets.Vision Res 16: 107–114, 1976a.

Hallett PE and Lightstone AD. Saccadic eye movements towards stimulitriggered by prior saccades. Vision Res 16: 99–106, 1976b.

Heide W, Blankenburg M, Zimmermann E, and Kompf D. Cortical controlof double-step saccades: implications for spatial orientation. Ann Neurol 38:739–748, 1995.

Helsinki. World Medical Organization: Declaration of Helsinki. Brit Med J313: 1448–1449, 1996.

Herter TM and Guitton D. Human head-free gaze saccades to targets flashedbefore gaze-pursuit are spatially accurate. J Neurophysiol 80: 2785–2789,1998.

Hodgson TL. The location marker effect. Saccadic latency increases withtarget eccentricity. Exp Brain Res 145: 539–542, 2002.

Jurgens R, Becker W, and Kornhuber HH. Natural and drug-inducedvariations of velocity and duration of human saccadic eye movements:evidence for a control of the neural pulse generator by local feedback. BiolCybern 39: 87–96, 1981.

Kalesnykas RP and Hallett PE. Retinal eccentricity and the latency of eyesaccades. Vision Res 34: 517–531, 1994.

Kanai R, Van Der Geest N, and Frens A. Inhibition of saccade initiation bypreceding smooth pursuit. Exp Brain Res 148: 300–307, 2003.

Keller EL, Gandhi NJ, and Weir PT. Discharge of superior collicularneurons during saccades made to moving targets. J Neurophysiol 76:3573–3577, 1996.

Keller E and Johnsen SD. Velocity prediction in corrective saccades duringsmooth-pursuit eye movements in monkey. Exp Brain Res 80: 525–531,1990.




Dow

nloaded from


Klein RM. Inhibition of return. Trends Cogn Sci 4: 138–147, 2000.Krauzlis RJ and Stone LS. Tracking with the mind’s eye. Trends Neurosci

22: 544–550, 1999.Leigh JR and Zee DS. The Neurology of Eye Movements. New York: Oxford,

1999.Lewis RF, Zee DS, Hayman MR, and Tamargo RJ. Oculomotor function in

the rhesus monkey after deafferentation of the extraocular muscles. ExpBrain Res 141: 349–358, 2001.

Lisberger SG, Morris EJ, and Tychsen L. Visual motion processing andsensory-motor integration for smooth pursuit eye movements. Annu RevNeurosci 10: 97–129, 1987.

May JG, Keller EL, and Suzuki DA. Smooth-pursuit eye movement deficitswith chemical lesions in the dorsolateral pontine nucleus of the monkey.J Neurophysiol 59: 952–977, 1988.

Maylor EA and Hockey R. Inhibitory component of externally controlledcovert orienting in visual space. J Exp Psychol Hum Percept Perform 11:777–787, 1985.

Mays LE and Sparks DL. Saccades are spatially, not retinocentrically, coded.Science 208: 1163–1165, 1980.

McKenzie A and Lisberger SG. Properties of signals that determine theamplitude and direction of saccadic eye movements in monkeys. J Neuro-physiol 56: 196–207, 1986.

Munoz DP and Wurtz RH. Role of the rostral superior colliculus in activevisual fixation and execution of express saccades. J Neurophysiol 67:1000–1002, 1992.

Mushiake H, Fujii N, and Tanji J. Microstimulation of the lateral wall of theintraparietal sulcus compared with the frontal eye field during oculomotortasks. J Neurophysiol 81: 1443–1448, 1999.

Niemann T and Hoffmann KP. Motion processing for saccadic eye move-ments during the visually induced sensation of ego-motion in humans.Vision Res 37: 3163–3170, 1997.

Ohtsuka K. Properties of memory-guided saccades toward targets flashedduring smooth pursuit in human subjects. Invest Ophthalmol Vis Sci 35:509–514, 1994.

Pare M and Wurtz RH. Monkey posterior parietal cortex neurons antidromi-cally activated from superior colliculus. J Neurophysiol 78: 3493–3497,1997.

Pierrot-Deseilligny C, Rivaud S, Gaymard B, and Agid Y. Cortical controlof memory-guided saccades in man. Exp Brain Res 83: 607–617, 1991.

Pola J and Wyatt HJ. Smooth pursuit: response characteristics, stimuli andmechanisms. In: Eye Movements: Vision and Visual Dysfunction, edited byCarpenter R. Houndmills, UK: Macmillan, 1991, p. 138–156.

Posner M, Rafal R, Choate L, and Vaughan J. Inhibition of return: neuralbasis and function. Cognit Neuropsychol 2: 211–228, 1985.

Rashbass C. The relationship between saccadic and smooth tracking eyemovements. J Physiol 159: 326–338, 1961.

Reddi BA, Asrress KN, and Carpenter RH. Accuracy, information, andresponse time in a saccadic decision task. J Neurophysiol 90: 3538–3546,2003.

Reddi BA and Carpenter RH. The influence of urgency on decision time. NatNeurosci 3: 827–830, 2000.

Reulen JP. Latency of visually evoked saccadic eye movements. I. Saccadiclatency and the facilitation model. Biol Cybern 50: 251–262, 1984a.

Reulen JP. Latency of visually evoked saccadic eye movements. II. Temporalproperties of the facilitation mechanism. Biol Cybern 50: 263–271, 1984b.

Robinson DA. A method of mesuring eye movement using a scleral searchcoil in a magnetic field. IEEE Trans Biomed Eng BME- 10: 137–145, 1963.

Robinson DA. Oculomotor control signals. In: Basic Mechanisms of OcularMotility and Their Clinical Implications, edited by Lennerstrand G andBach-y-Rita P. Oxford: Pergamon, 1975, p. 337–374.

Ron S, Vieville T, and Droulez J. Use of target velocity in saccadicprogramming. Brain Behav Evol 33: 85–89, 1989.

Saslow MG. Effects of components of displacement-step stimuli upon latencyfor saccadic eye movement. J Opt Soc Am 57: 1024–1029, 1967.

Schiller PH, Sandell JH, and Maunsell JH. The effect of frontal eye field andsuperior colliculus lesions on saccadic latencies in the rhesus monkey.J Neurophysiol 57: 1033–1049, 1987.

Schlag J and Schlag-Rey M. Colliding saccades may reveal the secret of theirmarching orders. Trends Neurosci 13: 410–415, 1990.

Schlag J, Schlag-Rey M, and Dassonville P. Interactions between natural andelectrically evoked saccades. II. At what time is eye position sampled as areference for the localization of a target? Exp Brain Res 76: 548–558, 1989.

Schlag J, Schlag-Rey M, and Dassonville P. Saccades can be aimed at thespatial location of targets flashed during pursuit. J Neurophysiol 64: 575–581, 1990.

Smeets JBJ and Bekkering H. Prediction of saccadic amplitude duringsmooth pursuit eye movements. Hum Move Sci 19: 275–295, 2000.

Sommer MA. The role of the thalamus in motor control. Curr Opin Neurobiol13: 663–670, 2003.

Sommer MA and Wurtz RH. A pathway in primate brain for internalmonitoring of movements. Science 296: 1480–1482, 2002.

Sommer MA and Wurtz RH. What the brain stem tells the frontal cortex. I.Oculomotor signals sent from superior colliculus to frontal eye field viamediodorsal thalamus. J Neurophysiol 91: 1381–1402, 2004a.

Sommer MA and Wurtz RH. What the brain stem tells the frontal cortex. II.Role of the SC-MD-FEF pathway in corollary discharge. J Neurophysiol 91:1403–1423, 2004b.

Stark L and Bridgeman B. Role of corollary discharge in space constancy.Percept Psychophys 34: 371–380, 1983.

Steinbach MJ. The palisade ending: an afferent source for eye positioninformation in humans. In: Advances in Strabismus Research: Basic andClinical Aspects, edited by Lennerstrand G and Ygge J. London: Portland,2000, p. 33–42.

Tanaka M, Yoshida T, and Fukushima K. Latency of saccades duringsmooth-pursuit eye movement in man. Directional asymmetries. Exp BrainRes 121: 92–98, 1998.

Thurston SE, Leigh RJ, Crawford T, Thompson A, and Kennard C. Twodistinct deficits of visual tracking caused by unilateral lesions of cerebralcortex in humans. Ann Neurol 23: 266–273, 1988.

Tian J, Schlag J, and Schlag-Rey M. Testing quasi-visual neurons in themonkey’s frontal eye field with the triple-step paradigm. Exp Brain Res 130:433–440, 2000.

Tobler PN, Felblinger J, Burki M, Nirkko AC, Ozdoba C, and Muri RM.Functional organisation of the saccadic reference system processing ex-traretinal signals in humans. Vision Res 41: 1351–1358, 2001.

Van Gisbergen JA, Robinson DA, and Gielen S. A quantitative analysis ofgeneration of saccadic eye movements by burst neurons. J Neurophysiol 45:417–442, 1981.

Weir CR. Spatial localisation: does extraocular muscle proprioception play arole? Graefes Arch Clin Exp Ophthalmol 238: 868–873, 2000.

Zivotofsky AZ, Rottach KG, Averbuch-Heller L, Kori AA, Thomas CW,Dell’Osso LF, and Leigh RJ. Saccades to remembered targets: the effectsof smooth pursuit and illusory stimulus motion. J Neurophysiol 76: 3617–3632, 1996.




Dow

nloaded from


blohm (2005) processing of retinal and extraretinal ...wexler.free.fr/library/files/blohm (2005)...

Documents